xref: /linux-6.15/include/linux/kvm_host.h (revision 75da60ff)
1 #ifndef __KVM_HOST_H
2 #define __KVM_HOST_H
3 
4 /*
5  * This work is licensed under the terms of the GNU GPL, version 2.  See
6  * the COPYING file in the top-level directory.
7  */
8 
9 #include <linux/types.h>
10 #include <linux/hardirq.h>
11 #include <linux/list.h>
12 #include <linux/mutex.h>
13 #include <linux/spinlock.h>
14 #include <linux/signal.h>
15 #include <linux/sched.h>
16 #include <linux/bug.h>
17 #include <linux/mm.h>
18 #include <linux/mmu_notifier.h>
19 #include <linux/preempt.h>
20 #include <linux/msi.h>
21 #include <linux/slab.h>
22 #include <linux/vmalloc.h>
23 #include <linux/rcupdate.h>
24 #include <linux/ratelimit.h>
25 #include <linux/err.h>
26 #include <linux/irqflags.h>
27 #include <linux/context_tracking.h>
28 #include <linux/irqbypass.h>
29 #include <linux/swait.h>
30 #include <linux/refcount.h>
31 #include <linux/nospec.h>
32 #include <asm/signal.h>
33 
34 #include <linux/kvm.h>
35 #include <linux/kvm_para.h>
36 
37 #include <linux/kvm_types.h>
38 
39 #include <asm/kvm_host.h>
40 
41 #ifndef KVM_MAX_VCPU_ID
42 #define KVM_MAX_VCPU_ID KVM_MAX_VCPUS
43 #endif
44 
45 /*
46  * The bit 16 ~ bit 31 of kvm_memory_region::flags are internally used
47  * in kvm, other bits are visible for userspace which are defined in
48  * include/linux/kvm_h.
49  */
50 #define KVM_MEMSLOT_INVALID	(1UL << 16)
51 
52 /*
53  * Bit 63 of the memslot generation number is an "update in-progress flag",
54  * e.g. is temporarily set for the duration of install_new_memslots().
55  * This flag effectively creates a unique generation number that is used to
56  * mark cached memslot data, e.g. MMIO accesses, as potentially being stale,
57  * i.e. may (or may not) have come from the previous memslots generation.
58  *
59  * This is necessary because the actual memslots update is not atomic with
60  * respect to the generation number update.  Updating the generation number
61  * first would allow a vCPU to cache a spte from the old memslots using the
62  * new generation number, and updating the generation number after switching
63  * to the new memslots would allow cache hits using the old generation number
64  * to reference the defunct memslots.
65  *
66  * This mechanism is used to prevent getting hits in KVM's caches while a
67  * memslot update is in-progress, and to prevent cache hits *after* updating
68  * the actual generation number against accesses that were inserted into the
69  * cache *before* the memslots were updated.
70  */
71 #define KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS	BIT_ULL(63)
72 
73 /* Two fragments for cross MMIO pages. */
74 #define KVM_MAX_MMIO_FRAGMENTS	2
75 
76 #ifndef KVM_ADDRESS_SPACE_NUM
77 #define KVM_ADDRESS_SPACE_NUM	1
78 #endif
79 
80 /*
81  * For the normal pfn, the highest 12 bits should be zero,
82  * so we can mask bit 62 ~ bit 52  to indicate the error pfn,
83  * mask bit 63 to indicate the noslot pfn.
84  */
85 #define KVM_PFN_ERR_MASK	(0x7ffULL << 52)
86 #define KVM_PFN_ERR_NOSLOT_MASK	(0xfffULL << 52)
87 #define KVM_PFN_NOSLOT		(0x1ULL << 63)
88 
89 #define KVM_PFN_ERR_FAULT	(KVM_PFN_ERR_MASK)
90 #define KVM_PFN_ERR_HWPOISON	(KVM_PFN_ERR_MASK + 1)
91 #define KVM_PFN_ERR_RO_FAULT	(KVM_PFN_ERR_MASK + 2)
92 
93 /*
94  * error pfns indicate that the gfn is in slot but faild to
95  * translate it to pfn on host.
96  */
97 static inline bool is_error_pfn(kvm_pfn_t pfn)
98 {
99 	return !!(pfn & KVM_PFN_ERR_MASK);
100 }
101 
102 /*
103  * error_noslot pfns indicate that the gfn can not be
104  * translated to pfn - it is not in slot or failed to
105  * translate it to pfn.
106  */
107 static inline bool is_error_noslot_pfn(kvm_pfn_t pfn)
108 {
109 	return !!(pfn & KVM_PFN_ERR_NOSLOT_MASK);
110 }
111 
112 /* noslot pfn indicates that the gfn is not in slot. */
113 static inline bool is_noslot_pfn(kvm_pfn_t pfn)
114 {
115 	return pfn == KVM_PFN_NOSLOT;
116 }
117 
118 /*
119  * architectures with KVM_HVA_ERR_BAD other than PAGE_OFFSET (e.g. s390)
120  * provide own defines and kvm_is_error_hva
121  */
122 #ifndef KVM_HVA_ERR_BAD
123 
124 #define KVM_HVA_ERR_BAD		(PAGE_OFFSET)
125 #define KVM_HVA_ERR_RO_BAD	(PAGE_OFFSET + PAGE_SIZE)
126 
127 static inline bool kvm_is_error_hva(unsigned long addr)
128 {
129 	return addr >= PAGE_OFFSET;
130 }
131 
132 #endif
133 
134 #define KVM_ERR_PTR_BAD_PAGE	(ERR_PTR(-ENOENT))
135 
136 static inline bool is_error_page(struct page *page)
137 {
138 	return IS_ERR(page);
139 }
140 
141 #define KVM_REQUEST_MASK           GENMASK(7,0)
142 #define KVM_REQUEST_NO_WAKEUP      BIT(8)
143 #define KVM_REQUEST_WAIT           BIT(9)
144 /*
145  * Architecture-independent vcpu->requests bit members
146  * Bits 4-7 are reserved for more arch-independent bits.
147  */
148 #define KVM_REQ_TLB_FLUSH         (0 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
149 #define KVM_REQ_MMU_RELOAD        (1 | KVM_REQUEST_WAIT | KVM_REQUEST_NO_WAKEUP)
150 #define KVM_REQ_PENDING_TIMER     2
151 #define KVM_REQ_UNHALT            3
152 #define KVM_REQUEST_ARCH_BASE     8
153 
154 #define KVM_ARCH_REQ_FLAGS(nr, flags) ({ \
155 	BUILD_BUG_ON((unsigned)(nr) >= (FIELD_SIZEOF(struct kvm_vcpu, requests) * 8) - KVM_REQUEST_ARCH_BASE); \
156 	(unsigned)(((nr) + KVM_REQUEST_ARCH_BASE) | (flags)); \
157 })
158 #define KVM_ARCH_REQ(nr)           KVM_ARCH_REQ_FLAGS(nr, 0)
159 
160 #define KVM_USERSPACE_IRQ_SOURCE_ID		0
161 #define KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID	1
162 
163 extern struct kmem_cache *kvm_vcpu_cache;
164 
165 extern spinlock_t kvm_lock;
166 extern struct list_head vm_list;
167 
168 struct kvm_io_range {
169 	gpa_t addr;
170 	int len;
171 	struct kvm_io_device *dev;
172 };
173 
174 #define NR_IOBUS_DEVS 1000
175 
176 struct kvm_io_bus {
177 	int dev_count;
178 	int ioeventfd_count;
179 	struct kvm_io_range range[];
180 };
181 
182 enum kvm_bus {
183 	KVM_MMIO_BUS,
184 	KVM_PIO_BUS,
185 	KVM_VIRTIO_CCW_NOTIFY_BUS,
186 	KVM_FAST_MMIO_BUS,
187 	KVM_NR_BUSES
188 };
189 
190 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
191 		     int len, const void *val);
192 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
193 			    gpa_t addr, int len, const void *val, long cookie);
194 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
195 		    int len, void *val);
196 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
197 			    int len, struct kvm_io_device *dev);
198 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
199 			       struct kvm_io_device *dev);
200 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
201 					 gpa_t addr);
202 
203 #ifdef CONFIG_KVM_ASYNC_PF
204 struct kvm_async_pf {
205 	struct work_struct work;
206 	struct list_head link;
207 	struct list_head queue;
208 	struct kvm_vcpu *vcpu;
209 	struct mm_struct *mm;
210 	gva_t gva;
211 	unsigned long addr;
212 	struct kvm_arch_async_pf arch;
213 	bool   wakeup_all;
214 };
215 
216 void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
217 void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu);
218 int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, unsigned long hva,
219 		       struct kvm_arch_async_pf *arch);
220 int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu);
221 #endif
222 
223 enum {
224 	OUTSIDE_GUEST_MODE,
225 	IN_GUEST_MODE,
226 	EXITING_GUEST_MODE,
227 	READING_SHADOW_PAGE_TABLES,
228 };
229 
230 /*
231  * Sometimes a large or cross-page mmio needs to be broken up into separate
232  * exits for userspace servicing.
233  */
234 struct kvm_mmio_fragment {
235 	gpa_t gpa;
236 	void *data;
237 	unsigned len;
238 };
239 
240 struct kvm_vcpu {
241 	struct kvm *kvm;
242 #ifdef CONFIG_PREEMPT_NOTIFIERS
243 	struct preempt_notifier preempt_notifier;
244 #endif
245 	int cpu;
246 	int vcpu_id;
247 	int srcu_idx;
248 	int mode;
249 	u64 requests;
250 	unsigned long guest_debug;
251 
252 	int pre_pcpu;
253 	struct list_head blocked_vcpu_list;
254 
255 	struct mutex mutex;
256 	struct kvm_run *run;
257 
258 	int guest_xcr0_loaded;
259 	struct swait_queue_head wq;
260 	struct pid __rcu *pid;
261 	int sigset_active;
262 	sigset_t sigset;
263 	struct kvm_vcpu_stat stat;
264 	unsigned int halt_poll_ns;
265 	bool valid_wakeup;
266 
267 #ifdef CONFIG_HAS_IOMEM
268 	int mmio_needed;
269 	int mmio_read_completed;
270 	int mmio_is_write;
271 	int mmio_cur_fragment;
272 	int mmio_nr_fragments;
273 	struct kvm_mmio_fragment mmio_fragments[KVM_MAX_MMIO_FRAGMENTS];
274 #endif
275 
276 #ifdef CONFIG_KVM_ASYNC_PF
277 	struct {
278 		u32 queued;
279 		struct list_head queue;
280 		struct list_head done;
281 		spinlock_t lock;
282 	} async_pf;
283 #endif
284 
285 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
286 	/*
287 	 * Cpu relax intercept or pause loop exit optimization
288 	 * in_spin_loop: set when a vcpu does a pause loop exit
289 	 *  or cpu relax intercepted.
290 	 * dy_eligible: indicates whether vcpu is eligible for directed yield.
291 	 */
292 	struct {
293 		bool in_spin_loop;
294 		bool dy_eligible;
295 	} spin_loop;
296 #endif
297 	bool preempted;
298 	struct kvm_vcpu_arch arch;
299 	struct dentry *debugfs_dentry;
300 };
301 
302 static inline int kvm_vcpu_exiting_guest_mode(struct kvm_vcpu *vcpu)
303 {
304 	/*
305 	 * The memory barrier ensures a previous write to vcpu->requests cannot
306 	 * be reordered with the read of vcpu->mode.  It pairs with the general
307 	 * memory barrier following the write of vcpu->mode in VCPU RUN.
308 	 */
309 	smp_mb__before_atomic();
310 	return cmpxchg(&vcpu->mode, IN_GUEST_MODE, EXITING_GUEST_MODE);
311 }
312 
313 /*
314  * Some of the bitops functions do not support too long bitmaps.
315  * This number must be determined not to exceed such limits.
316  */
317 #define KVM_MEM_MAX_NR_PAGES ((1UL << 31) - 1)
318 
319 struct kvm_memory_slot {
320 	gfn_t base_gfn;
321 	unsigned long npages;
322 	unsigned long *dirty_bitmap;
323 	struct kvm_arch_memory_slot arch;
324 	unsigned long userspace_addr;
325 	u32 flags;
326 	short id;
327 };
328 
329 static inline unsigned long kvm_dirty_bitmap_bytes(struct kvm_memory_slot *memslot)
330 {
331 	return ALIGN(memslot->npages, BITS_PER_LONG) / 8;
332 }
333 
334 static inline unsigned long *kvm_second_dirty_bitmap(struct kvm_memory_slot *memslot)
335 {
336 	unsigned long len = kvm_dirty_bitmap_bytes(memslot);
337 
338 	return memslot->dirty_bitmap + len / sizeof(*memslot->dirty_bitmap);
339 }
340 
341 struct kvm_s390_adapter_int {
342 	u64 ind_addr;
343 	u64 summary_addr;
344 	u64 ind_offset;
345 	u32 summary_offset;
346 	u32 adapter_id;
347 };
348 
349 struct kvm_hv_sint {
350 	u32 vcpu;
351 	u32 sint;
352 };
353 
354 struct kvm_kernel_irq_routing_entry {
355 	u32 gsi;
356 	u32 type;
357 	int (*set)(struct kvm_kernel_irq_routing_entry *e,
358 		   struct kvm *kvm, int irq_source_id, int level,
359 		   bool line_status);
360 	union {
361 		struct {
362 			unsigned irqchip;
363 			unsigned pin;
364 		} irqchip;
365 		struct {
366 			u32 address_lo;
367 			u32 address_hi;
368 			u32 data;
369 			u32 flags;
370 			u32 devid;
371 		} msi;
372 		struct kvm_s390_adapter_int adapter;
373 		struct kvm_hv_sint hv_sint;
374 	};
375 	struct hlist_node link;
376 };
377 
378 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
379 struct kvm_irq_routing_table {
380 	int chip[KVM_NR_IRQCHIPS][KVM_IRQCHIP_NUM_PINS];
381 	u32 nr_rt_entries;
382 	/*
383 	 * Array indexed by gsi. Each entry contains list of irq chips
384 	 * the gsi is connected to.
385 	 */
386 	struct hlist_head map[0];
387 };
388 #endif
389 
390 #ifndef KVM_PRIVATE_MEM_SLOTS
391 #define KVM_PRIVATE_MEM_SLOTS 0
392 #endif
393 
394 #ifndef KVM_MEM_SLOTS_NUM
395 #define KVM_MEM_SLOTS_NUM (KVM_USER_MEM_SLOTS + KVM_PRIVATE_MEM_SLOTS)
396 #endif
397 
398 #ifndef __KVM_VCPU_MULTIPLE_ADDRESS_SPACE
399 static inline int kvm_arch_vcpu_memslots_id(struct kvm_vcpu *vcpu)
400 {
401 	return 0;
402 }
403 #endif
404 
405 /*
406  * Note:
407  * memslots are not sorted by id anymore, please use id_to_memslot()
408  * to get the memslot by its id.
409  */
410 struct kvm_memslots {
411 	u64 generation;
412 	struct kvm_memory_slot memslots[KVM_MEM_SLOTS_NUM];
413 	/* The mapping table from slot id to the index in memslots[]. */
414 	short id_to_index[KVM_MEM_SLOTS_NUM];
415 	atomic_t lru_slot;
416 	int used_slots;
417 };
418 
419 struct kvm {
420 	spinlock_t mmu_lock;
421 	struct mutex slots_lock;
422 	struct mm_struct *mm; /* userspace tied to this vm */
423 	struct kvm_memslots __rcu *memslots[KVM_ADDRESS_SPACE_NUM];
424 	struct kvm_vcpu *vcpus[KVM_MAX_VCPUS];
425 
426 	/*
427 	 * created_vcpus is protected by kvm->lock, and is incremented
428 	 * at the beginning of KVM_CREATE_VCPU.  online_vcpus is only
429 	 * incremented after storing the kvm_vcpu pointer in vcpus,
430 	 * and is accessed atomically.
431 	 */
432 	atomic_t online_vcpus;
433 	int created_vcpus;
434 	int last_boosted_vcpu;
435 	struct list_head vm_list;
436 	struct mutex lock;
437 	struct kvm_io_bus __rcu *buses[KVM_NR_BUSES];
438 #ifdef CONFIG_HAVE_KVM_EVENTFD
439 	struct {
440 		spinlock_t        lock;
441 		struct list_head  items;
442 		struct list_head  resampler_list;
443 		struct mutex      resampler_lock;
444 	} irqfds;
445 	struct list_head ioeventfds;
446 #endif
447 	struct kvm_vm_stat stat;
448 	struct kvm_arch arch;
449 	refcount_t users_count;
450 #ifdef CONFIG_KVM_MMIO
451 	struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
452 	spinlock_t ring_lock;
453 	struct list_head coalesced_zones;
454 #endif
455 
456 	struct mutex irq_lock;
457 #ifdef CONFIG_HAVE_KVM_IRQCHIP
458 	/*
459 	 * Update side is protected by irq_lock.
460 	 */
461 	struct kvm_irq_routing_table __rcu *irq_routing;
462 #endif
463 #ifdef CONFIG_HAVE_KVM_IRQFD
464 	struct hlist_head irq_ack_notifier_list;
465 #endif
466 
467 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
468 	struct mmu_notifier mmu_notifier;
469 	unsigned long mmu_notifier_seq;
470 	long mmu_notifier_count;
471 #endif
472 	long tlbs_dirty;
473 	struct list_head devices;
474 	bool manual_dirty_log_protect;
475 	struct dentry *debugfs_dentry;
476 	struct kvm_stat_data **debugfs_stat_data;
477 	struct srcu_struct srcu;
478 	struct srcu_struct irq_srcu;
479 	pid_t userspace_pid;
480 };
481 
482 #define kvm_err(fmt, ...) \
483 	pr_err("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
484 #define kvm_info(fmt, ...) \
485 	pr_info("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
486 #define kvm_debug(fmt, ...) \
487 	pr_debug("kvm [%i]: " fmt, task_pid_nr(current), ## __VA_ARGS__)
488 #define kvm_debug_ratelimited(fmt, ...) \
489 	pr_debug_ratelimited("kvm [%i]: " fmt, task_pid_nr(current), \
490 			     ## __VA_ARGS__)
491 #define kvm_pr_unimpl(fmt, ...) \
492 	pr_err_ratelimited("kvm [%i]: " fmt, \
493 			   task_tgid_nr(current), ## __VA_ARGS__)
494 
495 /* The guest did something we don't support. */
496 #define vcpu_unimpl(vcpu, fmt, ...)					\
497 	kvm_pr_unimpl("vcpu%i, guest rIP: 0x%lx " fmt,			\
498 			(vcpu)->vcpu_id, kvm_rip_read(vcpu), ## __VA_ARGS__)
499 
500 #define vcpu_debug(vcpu, fmt, ...)					\
501 	kvm_debug("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
502 #define vcpu_debug_ratelimited(vcpu, fmt, ...)				\
503 	kvm_debug_ratelimited("vcpu%i " fmt, (vcpu)->vcpu_id,           \
504 			      ## __VA_ARGS__)
505 #define vcpu_err(vcpu, fmt, ...)					\
506 	kvm_err("vcpu%i " fmt, (vcpu)->vcpu_id, ## __VA_ARGS__)
507 
508 static inline struct kvm_io_bus *kvm_get_bus(struct kvm *kvm, enum kvm_bus idx)
509 {
510 	return srcu_dereference_check(kvm->buses[idx], &kvm->srcu,
511 				      lockdep_is_held(&kvm->slots_lock) ||
512 				      !refcount_read(&kvm->users_count));
513 }
514 
515 static inline struct kvm_vcpu *kvm_get_vcpu(struct kvm *kvm, int i)
516 {
517 	int num_vcpus = atomic_read(&kvm->online_vcpus);
518 	i = array_index_nospec(i, num_vcpus);
519 
520 	/* Pairs with smp_wmb() in kvm_vm_ioctl_create_vcpu.  */
521 	smp_rmb();
522 	return kvm->vcpus[i];
523 }
524 
525 #define kvm_for_each_vcpu(idx, vcpup, kvm) \
526 	for (idx = 0; \
527 	     idx < atomic_read(&kvm->online_vcpus) && \
528 	     (vcpup = kvm_get_vcpu(kvm, idx)) != NULL; \
529 	     idx++)
530 
531 static inline struct kvm_vcpu *kvm_get_vcpu_by_id(struct kvm *kvm, int id)
532 {
533 	struct kvm_vcpu *vcpu = NULL;
534 	int i;
535 
536 	if (id < 0)
537 		return NULL;
538 	if (id < KVM_MAX_VCPUS)
539 		vcpu = kvm_get_vcpu(kvm, id);
540 	if (vcpu && vcpu->vcpu_id == id)
541 		return vcpu;
542 	kvm_for_each_vcpu(i, vcpu, kvm)
543 		if (vcpu->vcpu_id == id)
544 			return vcpu;
545 	return NULL;
546 }
547 
548 static inline int kvm_vcpu_get_idx(struct kvm_vcpu *vcpu)
549 {
550 	struct kvm_vcpu *tmp;
551 	int idx;
552 
553 	kvm_for_each_vcpu(idx, tmp, vcpu->kvm)
554 		if (tmp == vcpu)
555 			return idx;
556 	BUG();
557 }
558 
559 #define kvm_for_each_memslot(memslot, slots)	\
560 	for (memslot = &slots->memslots[0];	\
561 	      memslot < slots->memslots + KVM_MEM_SLOTS_NUM && memslot->npages;\
562 		memslot++)
563 
564 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id);
565 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu);
566 
567 void vcpu_load(struct kvm_vcpu *vcpu);
568 void vcpu_put(struct kvm_vcpu *vcpu);
569 
570 #ifdef __KVM_HAVE_IOAPIC
571 void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm);
572 void kvm_arch_post_irq_routing_update(struct kvm *kvm);
573 #else
574 static inline void kvm_arch_post_irq_ack_notifier_list_update(struct kvm *kvm)
575 {
576 }
577 static inline void kvm_arch_post_irq_routing_update(struct kvm *kvm)
578 {
579 }
580 #endif
581 
582 #ifdef CONFIG_HAVE_KVM_IRQFD
583 int kvm_irqfd_init(void);
584 void kvm_irqfd_exit(void);
585 #else
586 static inline int kvm_irqfd_init(void)
587 {
588 	return 0;
589 }
590 
591 static inline void kvm_irqfd_exit(void)
592 {
593 }
594 #endif
595 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
596 		  struct module *module);
597 void kvm_exit(void);
598 
599 void kvm_get_kvm(struct kvm *kvm);
600 void kvm_put_kvm(struct kvm *kvm);
601 
602 static inline struct kvm_memslots *__kvm_memslots(struct kvm *kvm, int as_id)
603 {
604 	as_id = array_index_nospec(as_id, KVM_ADDRESS_SPACE_NUM);
605 	return srcu_dereference_check(kvm->memslots[as_id], &kvm->srcu,
606 			lockdep_is_held(&kvm->slots_lock) ||
607 			!refcount_read(&kvm->users_count));
608 }
609 
610 static inline struct kvm_memslots *kvm_memslots(struct kvm *kvm)
611 {
612 	return __kvm_memslots(kvm, 0);
613 }
614 
615 static inline struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu)
616 {
617 	int as_id = kvm_arch_vcpu_memslots_id(vcpu);
618 
619 	return __kvm_memslots(vcpu->kvm, as_id);
620 }
621 
622 static inline struct kvm_memory_slot *
623 id_to_memslot(struct kvm_memslots *slots, int id)
624 {
625 	int index = slots->id_to_index[id];
626 	struct kvm_memory_slot *slot;
627 
628 	slot = &slots->memslots[index];
629 
630 	WARN_ON(slot->id != id);
631 	return slot;
632 }
633 
634 /*
635  * KVM_SET_USER_MEMORY_REGION ioctl allows the following operations:
636  * - create a new memory slot
637  * - delete an existing memory slot
638  * - modify an existing memory slot
639  *   -- move it in the guest physical memory space
640  *   -- just change its flags
641  *
642  * Since flags can be changed by some of these operations, the following
643  * differentiation is the best we can do for __kvm_set_memory_region():
644  */
645 enum kvm_mr_change {
646 	KVM_MR_CREATE,
647 	KVM_MR_DELETE,
648 	KVM_MR_MOVE,
649 	KVM_MR_FLAGS_ONLY,
650 };
651 
652 int kvm_set_memory_region(struct kvm *kvm,
653 			  const struct kvm_userspace_memory_region *mem);
654 int __kvm_set_memory_region(struct kvm *kvm,
655 			    const struct kvm_userspace_memory_region *mem);
656 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
657 			   struct kvm_memory_slot *dont);
658 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
659 			    unsigned long npages);
660 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen);
661 int kvm_arch_prepare_memory_region(struct kvm *kvm,
662 				struct kvm_memory_slot *memslot,
663 				const struct kvm_userspace_memory_region *mem,
664 				enum kvm_mr_change change);
665 void kvm_arch_commit_memory_region(struct kvm *kvm,
666 				const struct kvm_userspace_memory_region *mem,
667 				const struct kvm_memory_slot *old,
668 				const struct kvm_memory_slot *new,
669 				enum kvm_mr_change change);
670 bool kvm_largepages_enabled(void);
671 void kvm_disable_largepages(void);
672 /* flush all memory translations */
673 void kvm_arch_flush_shadow_all(struct kvm *kvm);
674 /* flush memory translations pointing to 'slot' */
675 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
676 				   struct kvm_memory_slot *slot);
677 
678 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
679 			    struct page **pages, int nr_pages);
680 
681 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn);
682 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn);
683 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable);
684 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
685 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, gfn_t gfn,
686 				      bool *writable);
687 void kvm_release_page_clean(struct page *page);
688 void kvm_release_page_dirty(struct page *page);
689 void kvm_set_page_accessed(struct page *page);
690 
691 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn);
692 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn);
693 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
694 		      bool *writable);
695 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn);
696 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn);
697 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
698 			       bool atomic, bool *async, bool write_fault,
699 			       bool *writable);
700 
701 void kvm_release_pfn_clean(kvm_pfn_t pfn);
702 void kvm_release_pfn_dirty(kvm_pfn_t pfn);
703 void kvm_set_pfn_dirty(kvm_pfn_t pfn);
704 void kvm_set_pfn_accessed(kvm_pfn_t pfn);
705 void kvm_get_pfn(kvm_pfn_t pfn);
706 
707 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
708 			int len);
709 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
710 			  unsigned long len);
711 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len);
712 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
713 			   void *data, unsigned long len);
714 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
715 			 int offset, int len);
716 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
717 		    unsigned long len);
718 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
719 			   void *data, unsigned long len);
720 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
721 				  void *data, unsigned int offset,
722 				  unsigned long len);
723 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
724 			      gpa_t gpa, unsigned long len);
725 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len);
726 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len);
727 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn);
728 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn);
729 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn);
730 void mark_page_dirty(struct kvm *kvm, gfn_t gfn);
731 
732 struct kvm_memslots *kvm_vcpu_memslots(struct kvm_vcpu *vcpu);
733 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn);
734 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn);
735 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn);
736 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn);
737 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn);
738 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable);
739 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, int offset,
740 			     int len);
741 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
742 			       unsigned long len);
743 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data,
744 			unsigned long len);
745 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, const void *data,
746 			      int offset, int len);
747 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
748 			 unsigned long len);
749 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn);
750 
751 void kvm_sigset_activate(struct kvm_vcpu *vcpu);
752 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu);
753 
754 void kvm_vcpu_block(struct kvm_vcpu *vcpu);
755 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu);
756 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu);
757 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu);
758 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
759 int kvm_vcpu_yield_to(struct kvm_vcpu *target);
760 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu, bool usermode_vcpu_not_eligible);
761 
762 void kvm_flush_remote_tlbs(struct kvm *kvm);
763 void kvm_reload_remote_mmus(struct kvm *kvm);
764 
765 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
766 				 unsigned long *vcpu_bitmap, cpumask_var_t tmp);
767 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req);
768 
769 long kvm_arch_dev_ioctl(struct file *filp,
770 			unsigned int ioctl, unsigned long arg);
771 long kvm_arch_vcpu_ioctl(struct file *filp,
772 			 unsigned int ioctl, unsigned long arg);
773 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf);
774 
775 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext);
776 
777 int kvm_get_dirty_log(struct kvm *kvm,
778 			struct kvm_dirty_log *log, int *is_dirty);
779 
780 int kvm_get_dirty_log_protect(struct kvm *kvm,
781 			      struct kvm_dirty_log *log, bool *flush);
782 int kvm_clear_dirty_log_protect(struct kvm *kvm,
783 				struct kvm_clear_dirty_log *log, bool *flush);
784 
785 void kvm_arch_mmu_enable_log_dirty_pt_masked(struct kvm *kvm,
786 					struct kvm_memory_slot *slot,
787 					gfn_t gfn_offset,
788 					unsigned long mask);
789 
790 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
791 				struct kvm_dirty_log *log);
792 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
793 				  struct kvm_clear_dirty_log *log);
794 
795 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
796 			bool line_status);
797 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
798 			    struct kvm_enable_cap *cap);
799 long kvm_arch_vm_ioctl(struct file *filp,
800 		       unsigned int ioctl, unsigned long arg);
801 
802 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
803 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu);
804 
805 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
806 				    struct kvm_translation *tr);
807 
808 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
809 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs);
810 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
811 				  struct kvm_sregs *sregs);
812 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
813 				  struct kvm_sregs *sregs);
814 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
815 				    struct kvm_mp_state *mp_state);
816 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
817 				    struct kvm_mp_state *mp_state);
818 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
819 					struct kvm_guest_debug *dbg);
820 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run);
821 
822 int kvm_arch_init(void *opaque);
823 void kvm_arch_exit(void);
824 
825 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu);
826 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu);
827 
828 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu);
829 
830 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu);
831 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu);
832 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu);
833 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id);
834 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu);
835 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
836 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
837 
838 bool kvm_arch_has_vcpu_debugfs(void);
839 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu);
840 
841 int kvm_arch_hardware_enable(void);
842 void kvm_arch_hardware_disable(void);
843 int kvm_arch_hardware_setup(void);
844 void kvm_arch_hardware_unsetup(void);
845 void kvm_arch_check_processor_compat(void *rtn);
846 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu);
847 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu);
848 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu);
849 
850 #ifndef __KVM_HAVE_ARCH_VM_ALLOC
851 /*
852  * All architectures that want to use vzalloc currently also
853  * need their own kvm_arch_alloc_vm implementation.
854  */
855 static inline struct kvm *kvm_arch_alloc_vm(void)
856 {
857 	return kzalloc(sizeof(struct kvm), GFP_KERNEL);
858 }
859 
860 static inline void kvm_arch_free_vm(struct kvm *kvm)
861 {
862 	kfree(kvm);
863 }
864 #endif
865 
866 #ifndef __KVM_HAVE_ARCH_FLUSH_REMOTE_TLB
867 static inline int kvm_arch_flush_remote_tlb(struct kvm *kvm)
868 {
869 	return -ENOTSUPP;
870 }
871 #endif
872 
873 #ifdef __KVM_HAVE_ARCH_NONCOHERENT_DMA
874 void kvm_arch_register_noncoherent_dma(struct kvm *kvm);
875 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm);
876 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm);
877 #else
878 static inline void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
879 {
880 }
881 
882 static inline void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
883 {
884 }
885 
886 static inline bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
887 {
888 	return false;
889 }
890 #endif
891 #ifdef __KVM_HAVE_ARCH_ASSIGNED_DEVICE
892 void kvm_arch_start_assignment(struct kvm *kvm);
893 void kvm_arch_end_assignment(struct kvm *kvm);
894 bool kvm_arch_has_assigned_device(struct kvm *kvm);
895 #else
896 static inline void kvm_arch_start_assignment(struct kvm *kvm)
897 {
898 }
899 
900 static inline void kvm_arch_end_assignment(struct kvm *kvm)
901 {
902 }
903 
904 static inline bool kvm_arch_has_assigned_device(struct kvm *kvm)
905 {
906 	return false;
907 }
908 #endif
909 
910 static inline struct swait_queue_head *kvm_arch_vcpu_wq(struct kvm_vcpu *vcpu)
911 {
912 #ifdef __KVM_HAVE_ARCH_WQP
913 	return vcpu->arch.wqp;
914 #else
915 	return &vcpu->wq;
916 #endif
917 }
918 
919 #ifdef __KVM_HAVE_ARCH_INTC_INITIALIZED
920 /*
921  * returns true if the virtual interrupt controller is initialized and
922  * ready to accept virtual IRQ. On some architectures the virtual interrupt
923  * controller is dynamically instantiated and this is not always true.
924  */
925 bool kvm_arch_intc_initialized(struct kvm *kvm);
926 #else
927 static inline bool kvm_arch_intc_initialized(struct kvm *kvm)
928 {
929 	return true;
930 }
931 #endif
932 
933 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type);
934 void kvm_arch_destroy_vm(struct kvm *kvm);
935 void kvm_arch_sync_events(struct kvm *kvm);
936 
937 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu);
938 void kvm_vcpu_kick(struct kvm_vcpu *vcpu);
939 
940 bool kvm_is_reserved_pfn(kvm_pfn_t pfn);
941 
942 struct kvm_irq_ack_notifier {
943 	struct hlist_node link;
944 	unsigned gsi;
945 	void (*irq_acked)(struct kvm_irq_ack_notifier *kian);
946 };
947 
948 int kvm_irq_map_gsi(struct kvm *kvm,
949 		    struct kvm_kernel_irq_routing_entry *entries, int gsi);
950 int kvm_irq_map_chip_pin(struct kvm *kvm, unsigned irqchip, unsigned pin);
951 
952 int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level,
953 		bool line_status);
954 int kvm_set_msi(struct kvm_kernel_irq_routing_entry *irq_entry, struct kvm *kvm,
955 		int irq_source_id, int level, bool line_status);
956 int kvm_arch_set_irq_inatomic(struct kvm_kernel_irq_routing_entry *e,
957 			       struct kvm *kvm, int irq_source_id,
958 			       int level, bool line_status);
959 bool kvm_irq_has_notifier(struct kvm *kvm, unsigned irqchip, unsigned pin);
960 void kvm_notify_acked_gsi(struct kvm *kvm, int gsi);
961 void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin);
962 void kvm_register_irq_ack_notifier(struct kvm *kvm,
963 				   struct kvm_irq_ack_notifier *kian);
964 void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
965 				   struct kvm_irq_ack_notifier *kian);
966 int kvm_request_irq_source_id(struct kvm *kvm);
967 void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id);
968 
969 /*
970  * search_memslots() and __gfn_to_memslot() are here because they are
971  * used in non-modular code in arch/powerpc/kvm/book3s_hv_rm_mmu.c.
972  * gfn_to_memslot() itself isn't here as an inline because that would
973  * bloat other code too much.
974  */
975 static inline struct kvm_memory_slot *
976 search_memslots(struct kvm_memslots *slots, gfn_t gfn)
977 {
978 	int start = 0, end = slots->used_slots;
979 	int slot = atomic_read(&slots->lru_slot);
980 	struct kvm_memory_slot *memslots = slots->memslots;
981 
982 	if (gfn >= memslots[slot].base_gfn &&
983 	    gfn < memslots[slot].base_gfn + memslots[slot].npages)
984 		return &memslots[slot];
985 
986 	while (start < end) {
987 		slot = start + (end - start) / 2;
988 
989 		if (gfn >= memslots[slot].base_gfn)
990 			end = slot;
991 		else
992 			start = slot + 1;
993 	}
994 
995 	if (gfn >= memslots[start].base_gfn &&
996 	    gfn < memslots[start].base_gfn + memslots[start].npages) {
997 		atomic_set(&slots->lru_slot, start);
998 		return &memslots[start];
999 	}
1000 
1001 	return NULL;
1002 }
1003 
1004 static inline struct kvm_memory_slot *
1005 __gfn_to_memslot(struct kvm_memslots *slots, gfn_t gfn)
1006 {
1007 	return search_memslots(slots, gfn);
1008 }
1009 
1010 static inline unsigned long
1011 __gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1012 {
1013 	return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
1014 }
1015 
1016 static inline int memslot_id(struct kvm *kvm, gfn_t gfn)
1017 {
1018 	return gfn_to_memslot(kvm, gfn)->id;
1019 }
1020 
1021 static inline gfn_t
1022 hva_to_gfn_memslot(unsigned long hva, struct kvm_memory_slot *slot)
1023 {
1024 	gfn_t gfn_offset = (hva - slot->userspace_addr) >> PAGE_SHIFT;
1025 
1026 	return slot->base_gfn + gfn_offset;
1027 }
1028 
1029 static inline gpa_t gfn_to_gpa(gfn_t gfn)
1030 {
1031 	return (gpa_t)gfn << PAGE_SHIFT;
1032 }
1033 
1034 static inline gfn_t gpa_to_gfn(gpa_t gpa)
1035 {
1036 	return (gfn_t)(gpa >> PAGE_SHIFT);
1037 }
1038 
1039 static inline hpa_t pfn_to_hpa(kvm_pfn_t pfn)
1040 {
1041 	return (hpa_t)pfn << PAGE_SHIFT;
1042 }
1043 
1044 static inline struct page *kvm_vcpu_gpa_to_page(struct kvm_vcpu *vcpu,
1045 						gpa_t gpa)
1046 {
1047 	return kvm_vcpu_gfn_to_page(vcpu, gpa_to_gfn(gpa));
1048 }
1049 
1050 static inline bool kvm_is_error_gpa(struct kvm *kvm, gpa_t gpa)
1051 {
1052 	unsigned long hva = gfn_to_hva(kvm, gpa_to_gfn(gpa));
1053 
1054 	return kvm_is_error_hva(hva);
1055 }
1056 
1057 enum kvm_stat_kind {
1058 	KVM_STAT_VM,
1059 	KVM_STAT_VCPU,
1060 };
1061 
1062 struct kvm_stat_data {
1063 	int offset;
1064 	struct kvm *kvm;
1065 };
1066 
1067 struct kvm_stats_debugfs_item {
1068 	const char *name;
1069 	int offset;
1070 	enum kvm_stat_kind kind;
1071 };
1072 extern struct kvm_stats_debugfs_item debugfs_entries[];
1073 extern struct dentry *kvm_debugfs_dir;
1074 
1075 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1076 static inline int mmu_notifier_retry(struct kvm *kvm, unsigned long mmu_seq)
1077 {
1078 	if (unlikely(kvm->mmu_notifier_count))
1079 		return 1;
1080 	/*
1081 	 * Ensure the read of mmu_notifier_count happens before the read
1082 	 * of mmu_notifier_seq.  This interacts with the smp_wmb() in
1083 	 * mmu_notifier_invalidate_range_end to make sure that the caller
1084 	 * either sees the old (non-zero) value of mmu_notifier_count or
1085 	 * the new (incremented) value of mmu_notifier_seq.
1086 	 * PowerPC Book3s HV KVM calls this under a per-page lock
1087 	 * rather than under kvm->mmu_lock, for scalability, so
1088 	 * can't rely on kvm->mmu_lock to keep things ordered.
1089 	 */
1090 	smp_rmb();
1091 	if (kvm->mmu_notifier_seq != mmu_seq)
1092 		return 1;
1093 	return 0;
1094 }
1095 #endif
1096 
1097 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
1098 
1099 #define KVM_MAX_IRQ_ROUTES 4096 /* might need extension/rework in the future */
1100 
1101 bool kvm_arch_can_set_irq_routing(struct kvm *kvm);
1102 int kvm_set_irq_routing(struct kvm *kvm,
1103 			const struct kvm_irq_routing_entry *entries,
1104 			unsigned nr,
1105 			unsigned flags);
1106 int kvm_set_routing_entry(struct kvm *kvm,
1107 			  struct kvm_kernel_irq_routing_entry *e,
1108 			  const struct kvm_irq_routing_entry *ue);
1109 void kvm_free_irq_routing(struct kvm *kvm);
1110 
1111 #else
1112 
1113 static inline void kvm_free_irq_routing(struct kvm *kvm) {}
1114 
1115 #endif
1116 
1117 int kvm_send_userspace_msi(struct kvm *kvm, struct kvm_msi *msi);
1118 
1119 #ifdef CONFIG_HAVE_KVM_EVENTFD
1120 
1121 void kvm_eventfd_init(struct kvm *kvm);
1122 int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args);
1123 
1124 #ifdef CONFIG_HAVE_KVM_IRQFD
1125 int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args);
1126 void kvm_irqfd_release(struct kvm *kvm);
1127 void kvm_irq_routing_update(struct kvm *);
1128 #else
1129 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1130 {
1131 	return -EINVAL;
1132 }
1133 
1134 static inline void kvm_irqfd_release(struct kvm *kvm) {}
1135 #endif
1136 
1137 #else
1138 
1139 static inline void kvm_eventfd_init(struct kvm *kvm) {}
1140 
1141 static inline int kvm_irqfd(struct kvm *kvm, struct kvm_irqfd *args)
1142 {
1143 	return -EINVAL;
1144 }
1145 
1146 static inline void kvm_irqfd_release(struct kvm *kvm) {}
1147 
1148 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1149 static inline void kvm_irq_routing_update(struct kvm *kvm)
1150 {
1151 }
1152 #endif
1153 
1154 static inline int kvm_ioeventfd(struct kvm *kvm, struct kvm_ioeventfd *args)
1155 {
1156 	return -ENOSYS;
1157 }
1158 
1159 #endif /* CONFIG_HAVE_KVM_EVENTFD */
1160 
1161 void kvm_arch_irq_routing_update(struct kvm *kvm);
1162 
1163 static inline void kvm_make_request(int req, struct kvm_vcpu *vcpu)
1164 {
1165 	/*
1166 	 * Ensure the rest of the request is published to kvm_check_request's
1167 	 * caller.  Paired with the smp_mb__after_atomic in kvm_check_request.
1168 	 */
1169 	smp_wmb();
1170 	set_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1171 }
1172 
1173 static inline bool kvm_request_pending(struct kvm_vcpu *vcpu)
1174 {
1175 	return READ_ONCE(vcpu->requests);
1176 }
1177 
1178 static inline bool kvm_test_request(int req, struct kvm_vcpu *vcpu)
1179 {
1180 	return test_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1181 }
1182 
1183 static inline void kvm_clear_request(int req, struct kvm_vcpu *vcpu)
1184 {
1185 	clear_bit(req & KVM_REQUEST_MASK, (void *)&vcpu->requests);
1186 }
1187 
1188 static inline bool kvm_check_request(int req, struct kvm_vcpu *vcpu)
1189 {
1190 	if (kvm_test_request(req, vcpu)) {
1191 		kvm_clear_request(req, vcpu);
1192 
1193 		/*
1194 		 * Ensure the rest of the request is visible to kvm_check_request's
1195 		 * caller.  Paired with the smp_wmb in kvm_make_request.
1196 		 */
1197 		smp_mb__after_atomic();
1198 		return true;
1199 	} else {
1200 		return false;
1201 	}
1202 }
1203 
1204 extern bool kvm_rebooting;
1205 
1206 extern unsigned int halt_poll_ns;
1207 extern unsigned int halt_poll_ns_grow;
1208 extern unsigned int halt_poll_ns_grow_start;
1209 extern unsigned int halt_poll_ns_shrink;
1210 
1211 struct kvm_device {
1212 	struct kvm_device_ops *ops;
1213 	struct kvm *kvm;
1214 	void *private;
1215 	struct list_head vm_node;
1216 };
1217 
1218 /* create, destroy, and name are mandatory */
1219 struct kvm_device_ops {
1220 	const char *name;
1221 
1222 	/*
1223 	 * create is called holding kvm->lock and any operations not suitable
1224 	 * to do while holding the lock should be deferred to init (see
1225 	 * below).
1226 	 */
1227 	int (*create)(struct kvm_device *dev, u32 type);
1228 
1229 	/*
1230 	 * init is called after create if create is successful and is called
1231 	 * outside of holding kvm->lock.
1232 	 */
1233 	void (*init)(struct kvm_device *dev);
1234 
1235 	/*
1236 	 * Destroy is responsible for freeing dev.
1237 	 *
1238 	 * Destroy may be called before or after destructors are called
1239 	 * on emulated I/O regions, depending on whether a reference is
1240 	 * held by a vcpu or other kvm component that gets destroyed
1241 	 * after the emulated I/O.
1242 	 */
1243 	void (*destroy)(struct kvm_device *dev);
1244 
1245 	int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1246 	int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1247 	int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
1248 	long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
1249 		      unsigned long arg);
1250 };
1251 
1252 void kvm_device_get(struct kvm_device *dev);
1253 void kvm_device_put(struct kvm_device *dev);
1254 struct kvm_device *kvm_device_from_filp(struct file *filp);
1255 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type);
1256 void kvm_unregister_device_ops(u32 type);
1257 
1258 extern struct kvm_device_ops kvm_mpic_ops;
1259 extern struct kvm_device_ops kvm_arm_vgic_v2_ops;
1260 extern struct kvm_device_ops kvm_arm_vgic_v3_ops;
1261 
1262 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1263 
1264 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
1265 {
1266 	vcpu->spin_loop.in_spin_loop = val;
1267 }
1268 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
1269 {
1270 	vcpu->spin_loop.dy_eligible = val;
1271 }
1272 
1273 #else /* !CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
1274 
1275 static inline void kvm_vcpu_set_in_spin_loop(struct kvm_vcpu *vcpu, bool val)
1276 {
1277 }
1278 
1279 static inline void kvm_vcpu_set_dy_eligible(struct kvm_vcpu *vcpu, bool val)
1280 {
1281 }
1282 #endif /* CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT */
1283 
1284 #ifdef CONFIG_HAVE_KVM_IRQ_BYPASS
1285 bool kvm_arch_has_irq_bypass(void);
1286 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *,
1287 			   struct irq_bypass_producer *);
1288 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *,
1289 			   struct irq_bypass_producer *);
1290 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *);
1291 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *);
1292 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
1293 				  uint32_t guest_irq, bool set);
1294 #endif /* CONFIG_HAVE_KVM_IRQ_BYPASS */
1295 
1296 #ifdef CONFIG_HAVE_KVM_INVALID_WAKEUPS
1297 /* If we wakeup during the poll time, was it a sucessful poll? */
1298 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
1299 {
1300 	return vcpu->valid_wakeup;
1301 }
1302 
1303 #else
1304 static inline bool vcpu_valid_wakeup(struct kvm_vcpu *vcpu)
1305 {
1306 	return true;
1307 }
1308 #endif /* CONFIG_HAVE_KVM_INVALID_WAKEUPS */
1309 
1310 #ifdef CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL
1311 long kvm_arch_vcpu_async_ioctl(struct file *filp,
1312 			       unsigned int ioctl, unsigned long arg);
1313 #else
1314 static inline long kvm_arch_vcpu_async_ioctl(struct file *filp,
1315 					     unsigned int ioctl,
1316 					     unsigned long arg)
1317 {
1318 	return -ENOIOCTLCMD;
1319 }
1320 #endif /* CONFIG_HAVE_KVM_VCPU_ASYNC_IOCTL */
1321 
1322 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
1323 		unsigned long start, unsigned long end, bool blockable);
1324 
1325 #ifdef CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE
1326 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu);
1327 #else
1328 static inline int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
1329 {
1330 	return 0;
1331 }
1332 #endif /* CONFIG_HAVE_KVM_VCPU_RUN_PID_CHANGE */
1333 
1334 #endif
1335